����ɫ��

It’s colourless, odourless and is integral to the world’s energy supply.

Natural gas is extracted and transported to every corner of our society to heat homes and water, cook, dry clothes, and operate refrigeration and cooling equipment, among many other uses.

One of the biggest challenges remains detecting, localizing and quantifying leaks, which flies in the face of attempts to reduce greenhouse gas (GHG) emissions.

Schulich School of Engineering researcher , PhD, P.Eng, is hoping to change that by using infrared imaging to quantify natural gas leaks.

“Natural gas leaks are invisible to the naked eye and to regular cameras, but are visible to infrared cameras,” Du says. “So it would be possible to abstract quantitative information from those images.”

Recently, Du received a $240,000 grant through the to help continue his research.

An imperfect approach

Currently, industry uses a mobile methane detector or infrared camera to identity and locate the leak, then uses a high-flow meter to measure the emission rate.

Du says this approach is labour intensive, it’s not in real time, and it may miss smaller leaking points.

“It would be highly desirable if the emission rate can also be quantified by infrared imaging to make leak monitoring continuous, automatic, and real time,” he says. “However, the appearance of natural gas plumes in infrared imaging is affected by many factors, including flow rate, distance, background and dispersion behaviour.”

Thanks to the rapid development of machine learning techniques and high computing speed of microcomputers, Du says he plans to create a smarter detecting process by adding artificial intelligence, or as he calls it, a “brain,” to the camera, making it more realistic to identify and quantify leaks more quickly and more accurately.

Systematic improvements

Du is taking a three-pronged approach to his study, as he aims to create a system that is not only accurate but also proactive.

It starts with plume detection, using different mediums and analyses to isolate plumes from their backgrounds and characterizing factors relating to the leak’s flow rate.

From there, Du will use machine learning to construct synthetic parameters, before moving on to building algorithm structures and testing.

“We hope to build a data-driven model for quantifying natural gas emission rates from the infrared video images so that the system can be used to continuously monitor leaks in the field of view,” he says. “Such a system can help with locating, quantifying and early alarming of major leaks so that timely intervention can happen to reduce greenhouse gas emissions.”

A sustainable future for all

Du, who received funding, is grateful for the opportunity to make a difference in the world.

“Although infrared cameras have been used for leak detection by many facility operators, there have been no systematic studies on their effectiveness for quantifying leaks,” he says. “Integrating traditional techniques, augmented with the introduction of novel machine-learning analysis, will allow for a new perspective and open doors for early and comprehensive detection and quantification of leaks in the future.”

A total of 102 research projects across Canada share in the $45 million announced in April by the federal government.